Thin-film lithium niobate is a sub-micron single-crystal Pockels-effect layer bonded onto silicon or oxide wafers, and in 2026 it moved from lab modulators to 200 mm wafer-scale integration with >95% printing yield and 320 Gb/s unamplified links, while simultaneously setting records in integrated quantum optics such as 18 dB of on-chip squeezing ref.
Summary
Lithium niobate (LiNbO3) has been the workhouse crystal of telecoms modulators for decades, but as a bulk material it is used in centimetre-scale discrete components. Thin-film lithium niobate (TFLN) is the same crystal bonded as a sub-micron layer onto an insulator and etched into ridge waveguides, so that light is confined tightly enough for the electrodes to sit microns away from the optical mode. The physics being exploited is the Pockels effect: an applied voltage changes the refractive index linearly and essentially instantaneously, with no free-carrier injection. That gives modulators with flat, high-frequency response, low chirp and low optical loss, in contrast to silicon modulators which rely on carrier depletion. The same crystal has a large second-order nonlinearity (used for frequency conversion and photon-pair generation when the domains are periodically poled) and is piezoelectric (used for acousto-optic devices), so one material covers modulation, nonlinear optics and transduction.
The parameters that decide whether TFLN wins are not exotic. First, the voltage-length product VπL, which sets how short and how low-power a modulator can be: recent hybrid arrays report 2.52-2.68 V·cm with half-wave voltages of 3.60-3.83 V and >40 GHz 3 dB bandwidth. Second, insertion loss, and specifically fibre-to-chip and laser-to-chip coupling: the same array reports 15.19-16.55 dB bare-chip insertion loss plus additional DFB bonding loss, which is the sort of number that keeps a device in the lab. Third, manufacturability at silicon-photonics wafer scale, since LN is not a CMOS-line material and must be bonded or transfer-printed. Fourth, bias stability: DC drift from charge redistribution and surface screening makes even the characterisation of high-Q LN cavities unreliable under quasi-static voltage scans.
The 2026 evidence base splits into two distinct application stacks. In datacentre interconnect, micro-transfer printing has been demonstrated across four full 200 mm wafers with 3σ placement accuracy down to 420 nm, printing yield above 95%, insertion loss below 2 dB across 600 phase modulators and a 4 V half-wave voltage, with a variability-aware design workflow built on pilot-line fabrication data to make that repeatable, and a system demonstration of 320 Gb/s unamplified transmission pairing a 100 GHz germanium photodiode with a TFLN Mach-Zehnder modulator on a foundry-compatible silicon photonics platform ref. In quantum and nonlinear photonics, TFLN is producing the strongest integrated results anywhere: 18 dB of continuous-wave on-chip squeezing in a 1.6 cm adaptively poled waveguide, described as the highest for any integrated photonic platform; the first on-chip telecom quantum memory in TFLN using erbium doping; and cavity-enhanced memory with 23.3% on-chip storage efficiency plus electro-optic frequency routing at 20 MHz. Poling length, historically capped near 10 mm, has been pushed to 70 mm of continuous poled area at a 3 µm period.
Evidence base
Analogue RF: a third application stack (added 15 Oct 2026)
Open questions
- Can fibre-to-chip and laser-to-chip coupling losses be cut from the 15-16 dB bare-chip figures reported in hybrid TFLN transmitter arrays to link-budget-viable levels, and at what packaging cost?
- Is DC drift a material property that must be designed around at circuit level, or can buffer-layer and surface engineering eliminate it well enough for multi-year biased operation?
- Does the >95% printing yield and 420 nm placement accuracy hold across many wafer lots and outside the originating pilot line?
- How far does thin-film lithium tantalate encroach beyond the visible band, given its demonstrated advantage in sustained optical power handling at 532 nm?
- Is TFLN the only practical material for on-chip Brillouin RF filtering, or does indium phosphide (now showing large backward Brillouin gain on a foundry platform) match it at lower cost?
Assessment drafted 2026-08-31 from up to 18 KB sources using the technology-scorecard framework; scores are a draft read pending review.